To register a custom RWARE environment, go to epymarl-rware/src/marl.py file and give the layout of the custom environment. To register the environment just run marl.py file.
To train a RL algorithm on the environment use the command.
python src/main.py --config=qmix --env-config=gymma with env_args.time_limit=500 env_args.key="marl:your-env-name" save_model=True
If common reward is insufficient and individual rewards are needed, common_reward=False
.
python src/main.py --config=qmix --env-config=gymma with env_args.time_limit=500 env_args.key="marl:your-env-name" save_model=True common_reward=False
To visualize results, run from the root folder, after selecting the required metric.
python3 plot_results.py --path ./results/sacred/qmix/your-env-name --metric test_return_mean --save_dir ./plots
To render results, run from the root folder, after selecting the required model.
python src/main.py --config=qmix --env-config=gymma with env_args.time_limit=25 env_args.key="marl:r-tiny-2ag-v2" checkpoint_path="./results/models/your-model-folder" evaluate=True render=True
For readme of EpyMARL, visit https://github.com/uoe-agents/epymarl/blob/main/README.md